Nobody gets the question right first time. A chat holds what has already been established, so refining an answer costs a sentence rather than a rewrite.
Exploratory analysis is a sequence of small corrections. You ask something roughly right, look at the shape of the answer — literally, because the answer is drawn as a chart — and adjust. In a query box every adjustment means editing SQL you just read; in a chat it means a clause. "Break that down by plan." "Drop the internal accounts." "Same thing for last year." The tables, joins and filters already agreed on stay agreed on.
The risk in a stateful interface is that you lose track of what is still applied — a filter set four questions ago quietly shaping an answer you are about to take to a meeting. That is why every answer arrives with its full SQL rather than a diff. The query is the state, written out, and reading it takes a second.
The query you worked out in March is still there in September, and running it again gives you September's numbers rather than March's cached ones. In practice a saved conversation becomes a lightweight report: the reasoning, the query and the current answer in one place, without anyone having to promote it into a dashboard.
Two plausible date columns, or two tables that both look like the orders table, is ambiguity the schema cannot settle. It comes back as a question rather than a confident guess — and once you have answered it, you can write it down as a rule so the same question is not asked next month.
More answers on the main FAQ, or ask us directly.
PostgreSQL
MySQL
For analysis
Every page in this section covers a different question about the same product. See all solutions.
Start free, upgrade when you are asking enough questions for it to matter.
